The key bit that companies that shift hard to software seem to figure out:
How many of those corner cases are actually worth bothering with? Is the net loss associated with a corner case greater than the significant gain of dropping an FTE or FTEs?
Sometimes it is (safety related things, etc.), but many times, it's not...just because you CAN handle something doesn't mean it's worth it do so.
It’s something the organization needs to decide. If you automate a process it means you are setting it in stone and lose flexibility . Maybe that’s a good thing or not but something people need to be aware of and accept.